Quantifying Memory Use in Reinforcement Learning with Temporal Range